Implement S3 storage class for Streaming and External basic transfer adapters

[tomeksabala]

Implements a feature described in issue #3

I have created S3 storage with class giftless.storage.aws_s3.AwsS3Storage implementing two interfaces: StreamingStorage and ExternalStorage.

I have decided to support AWS authentication in a native way to boto3. In my opinion AWS users are used to the most common ways to provide credentials as I've explained in docs/source/storage-backends.md:127

One outstanding issue is the requirement.txt file in this PR. I needed to add boto3 and botocore requirements to requirements.in. After running make requirements.txt I couldn't keep the old formatting in place. I can't fix it once I'm told how to do this properly.

[shevron]

I think this is great! I have a few comments, I don't think any of them is huge, but would be good to fix or get your feedback on them. I think the requirements.txt formatting is a result of an older version of pip-tools - try to upgrade to the latest and see what happens.

[tomeksabala]

Thanks @shevron . I've fixed/updated all listed issues. I have left two conversations unresolved to get your feedback on:

1. ### Running updated yaml config with uWSGI section in the readme
2. Type hints around safe_filename call.

[shevron]

Getting very close now 👍

[shevron]

I think it would be nicer to not set content-disposition unless we need to (that is unless the user specified a custom name).

Something like:

Suggested change

	raw_filename = extra.get('filename') if extra else oid
	assert raw_filename
	filename = safe_filename(raw_filename)
	params = {
	'Bucket': self.bucket_name,
	'Key': self._get_blob_path(prefix, oid),
	'ResponseContentDisposition': f"attachment;filename={filename}"
	}
	params = {'Bucket': self.bucket_name, 'Key': self._get_blob_path(prefix, oid)}
	if extra and 'filename' in extra:
	filename = safe_filename(extra['filename'])
	params['ResponseContentDisposition'] = f'attachment; filename="{filename}"'

I think this will also pass type checking.
I've also added double quotes around the filename in the header, it is not strictly required unless the filename contains a space or a semicolon IIRC. While right now safe_filename prevents that, in the future I'd like to make it less strict and then it will be required.

[tomeksabala]

That's helpful, thank you. I have manually verified and actually the 'ResponseContentDisposition' can only be provided if we want a custom filename for downloaded file. Otherwise it defaults to object it oid.
Type checks also passed this change. Thank you for fixing the syntax of the header too.
I have changed only the params dict init so it aligns better with the rest of the dicts in the class.

[tomeksabala]

This is great feedback. Thanks as I've learned something new because of it :)

[shevron]

Looks very good, thanks.

[shevron]

@tomeksabala this looks great now, I'm merging this to master, and will issue a new release with this included in the next few days - I want to ensure the docs are up to date, and run the full test suite with a live AWS account before I make an official "stable" release.

[shevron]

/cc @rufuspollock

[rufuspollock]

@tomeksabala @shevron 👏 👏